What is DoShiCo?
DoShiCo is a benchmark to compare the performance of different deep neural network policies on the task of monocular collision avoidance. Between the simulated and the real world is a very large domain shift. Dealing with this domain shift is crucial for making deep neural policies perform well in the real world. It is however very difficult to compare different training methods or architectures as long as there is no clear benchmark. DoShiCo responds to this demand.
Try it yourself!
In order to use DoShiCo as a benchmark we made everything accessible. In the following steps we explain first how to get the general required software packages. The second step describes the comunication with the drone either in simulator or in the real world. The final step explains how you can reproduce our results published in our paper.